Ohanafy is heading to NBWA 2026Meet us
TECHNICAL DEEP-DIVE

How Ohanafy AI works.

The architecture. The models. The guardrails. The reason your data stays yours.

Ohanafy · Dashboard
Live
Revenue
$3.42M
Mar · MTD
Routes
8
Today · live
Fill rate
99.2%
+0.8 wow
Active reps
47
of 52
Depletion · Q1
On pace
Depletion alert · 92%

The architecture

Three layers, stacked. Customer data sits at the bottom and never leaves Salesforce. Models orchestrate in the middle. Agents act on top.

OHANAFY AI AGENTSAnomaly detectionForecastingActionAI ORCHESTRATIONClaude · LLMSalesforce EinsteinCustom Ohanafy modelsSALESFORCE PLATFORMCustomer data · Permissions · Audit

Three layers. One platform. Customer data lives at the bottom and never leaves.

The models

No single model wins everything. Ohanafy AI orchestrates three model classes, each tuned to what they do best.

Claude (Anthropic)

Used for: natural language understanding, email drafting, executive summary generation, complex reasoning.

Why: best-in-class for following instructions and staying in your voice.

Salesforce Einstein

Used for: account scoring, predictive forecasting, classification tasks.

Why: native to your data model, no data egress.

Custom Ohanafy models

Used for: anomaly detection on depletions, pricing pattern analysis, route optimization.

Why: trained specifically on beverage industry data.

Your data, your rules.

Customer data never leaves the customer's Salesforce org for training. AI inference happens with scoped data passed in-context, not added to training sets. Your records stay where Salesforce hosts them.

Customer-owned and customer-deletable. Audit logs on every AI action capture the prompt, the response, and the reviewer. If you want a record gone, it goes.

Inheritance from the Salesforce platform handles the heavy lifting: row-level security, field-level encryption, MFA, SSO, audit trails. AI runs inside that perimeter, not around it.

0
customer data used to train base models
100%
audit logs on AI actions
Native
Salesforce security inheritance
Yours
customer-controlled retention

The guardrails

Four hard rules. Built into every workflow.

Action authorization

AI never takes irreversible action without explicit user authorization. Drafts emails. Suggests orders. Humans approve.

Auditability

Every AI action is logged with the prompt, the response, and the human review state.

Confidence thresholds

AI surfaces predictions with confidence intervals. Below threshold means human review required.

Human-in-the-loop

AI is a senior analyst, not the decision-maker. Final calls always sit with humans.

How fast and what it costs

PERFORMANCE
  • Median responseUnder 2 seconds
  • ThroughputHandles peak load in every customer org
  • UptimeInherits Salesforce SLA
COST TRANSPARENCY
  • Included in licenseYes
  • Per-action feesNone
  • Token billingNone

What is shipping next

  • Demand planning
  • Territory optimization
  • Promotional ROI
  • Competitive pricing intelligence
  • Inventory rebalancing
  • Customer segmentation

Quarterly roadmap reviewed with customers. Email product@ohanafy.com to influence priority.

TRY IT YOURSELF

The playground is the proof.

The architecture is one thing. Asking your own question and getting an answer is another. Open the playground and try a real prompt.